Anthropic Ownership: Amazon and Google Stakes and Control

Amazon and Google are Anthropic’s largest outside economic owners and its largest cloud counterparties, but the October 26, 2026 S-1 walls them off from formal control: their stock is non-voting Class N, contractual ownership caps sit well below a control threshold, and neither has a board seat or observer right. The real leverage is commercial, not corporate.

Control is deliberately separated from capital

Anthropic’s post-IPO structure treats Amazon and Google as funders and suppliers, not governors. Immediately before the offering, the company issues one Class F share to a Founder LLC whose initial members are the seven co-founders. That single share carries a fixed 50.1% of the vote on matters submitted to stockholders generally, including election of directors other than the Class T directors, and it has no economic rights. Class A holders together are capped at 49.9%. A separate Class T share, held by the Long-Term Benefit Trust, elects a majority of the board. Class N—the class “primarily” held by strategic partners—has the same economics as Class A but no vote except where law requires one, and it converts into voting Class A only on a sale or transfer to an unaffiliated third party.

The S-1 states that after preferred stock and convertible notes convert, Amazon and Google will each hold greater than 5% of outstanding capital stock, in non-voting Class N, and that those holdings “represent a significant economic interest.” Exact share counts and percentages are blank in the filing. Because the Trust can elect a board majority, Anthropic qualifies as a Nasdaq “controlled company.” It says it expects to use that exemption only for the nominating and corporate governance committee.

What this means for anyone trying to buy influence: writing a check does not buy a vote, a director, or a veto. The only path from Class N to voting stock is a transfer out of the Amazon or Google corporate family.

Ownership stakes are large, capped, and still not fully disclosed

The hard numbers in the S-1 are dollars and conversion mechanics, not percentages. Public estimates of the stakes disagree because later rounds diluted earlier holdings and because conversion is gated by caps the holders can waive.

Amazon. TechCrunch and Bloomberg both put cumulative cash invested at about $18 billion: roughly $8 billion of convertible notes from 2023 through 2025, plus $5 billion in each of the Series G and Series H preferred rounds in 2026. As of June 30, 2026, $6.2 billion of note principal remained outstanding and could convert into as many as 1.84 billion Class N shares (conversion prices cited in secondary analyses at about $2.09 and $5.05). Conversion stops once Amazon would exceed a 12% ownership cap on a partially diluted basis, unless Amazon waives the cap. Excess stays as non-interest-bearing rollover notes that convert in later quarters only as room opens under the same limit. An April 2026 facility still allows up to $15 billion more of non-voting stock or notes, drawn against compute milestones. On Amazon’s own books at June 30, non-voting preferred was carried at $92.5 billion and the notes at an estimated fair value of $97.9 billion.

Bloomberg’s reading of the filing puts Amazon’s current economic stake at roughly 9% on an as-converted, fully diluted basis, with the notes theoretically worth another ~11%—but the 12% ceiling binds unless waived. PitchBook has said full conversion without the cap would imply roughly one-fifth. Those figures are calculations, not S-1 disclosures.

Google. Court filings reported by The New York Times in 2025 put Google at about 14% on a fully diluted basis, contractually capped near 15%, with no voting rights, board seats, or board-observer rights. The 2026 S-1 shows further investment: about $10 billion of Series G-2 in March 2026 and $10 billion of Series H in May 2026, against a March 31, 2026 note-purchase agreement for up to $30 billion more through December 31, 2030, tied to operational and revenue milestones. After the Series H check, about $20 billion of that facility remained. Bloomberg calculates that Google has invested about $23 billion in total and holds an economic stake of at least 6.6% on an as-converted, fully diluted basis—excluding a separate $300 million investment in 2022 that the S-1 does not detail. A PitchBook analyst told TechCrunch the stake is closer to 8% after dilution from the 14% court-filing figure. The 2026 agreement caps ownership at 14.99% unless Google waives it; excess is funded through a zero-interest rollover note. Google also has an option to buy up to $10 billion of Class N at the IPO price in a concurrent private placement.

UK Competition and Markets Authority materials from the earlier Google partnership review described the stake as non-voting, with no board representation, and found that consultation rights did not amount to material influence sufficient for a merger finding. That is the design: stay under control thresholds that would turn a strategic investment into an acquisition review.

For a competitor or a public buyer of ANT, the relevant risk is not a sudden Amazon or Google takeover vote. It is that both can still choose to waive their caps, and that Class N becomes ordinary float the moment either sells to an outsider—so the “permanent” strategic capital is also future supply.

The commercial contracts are the real governance

The S-1’s binding constraints on Anthropic are multi-year, largely non-cancellable compute minimums and a distribution channel that already moves nearly half of revenue.

• Anthropic has committed to spend at least $110 billion with Amazon from May 2026 through April 2036, and at least $114.5 billion with Google from April 2026 through July 2033. Both are owed regardless of actual usage; shortfalls are generally trued up. Neither deal is cancellable except for an uncured material breach by the provider. AWS is described as the primary training and cloud provider for mission-critical workloads and supplies Trainium capacity, including the Project Rainier cluster. Google supplies Cloud capacity and TPUs; Anthropic has also estimated another $125–150 billion of additional TPU capacity from 2028 to 2031, outside the minimum table.
• Distribution is the second loop. Confidential prospectus figures reported by Reuters, and the public S-1, show sales through Amazon and Google cloud marketplaces at 47% of 2025 revenue (about $2.16 billion) and about 48% of first-half 2026 revenue (Amazon, Google, and Microsoft marketplaces together in the later period). Reuters’ analysis of the confidential filing put distribution fees at roughly $351 million in 2025, about 16 cents on each marketplace dollar. Those platforms also collected a rising share of customer receivables—60% of $909 million outstanding at the end of 2025.
• Chip exposure runs through Google’s design even when the lessor is someone else. The largest single equipment commitment in the filing is about $161.2 billion of Broadcom leases on TPU-based hardware. Broadcom can be issued up to $42 billion of notes solely to finance that rent.

The mechanism is circular by construction. Amazon and Google put capital in, often convertible into more Class N. Anthropic spends a contracted minimum back to them for compute whether or not usage matches the forecast. They also resell Claude, keep a fee, and see pricing and commercial terms. Their own models—Gemini at Google, Nova and related offerings at Amazon—compete with Claude on the same platforms.

How the IPO filing describes the conflict—and what it leaves out of “related party”

The S-1 does not hide the tangle. It describes Amazon, Google, and Microsoft as “investors, customers, distribution partners and competitors” at once. It warns that reliance on a limited set of cloud providers and suppliers, some of which are also strategic investors and customers, “may create complex situations that could lead to conflicts of interest and adversely affect our ability to secure computing power,” and that the relationships “create potential conflicts of interest that are difficult to fully anticipate or mitigate.” Specific channels of harm named or closely paraphrased in coverage of the filing:

• The cloud companies have “visibility into our pricing and commercial terms,” which can shape how they allocate capacity and how hard they sell Claude versus their own models.
• They can prioritize capacity for their own AI products or slow delivery.
• As shareholders they have a financial stake in Anthropic’s valuation, which can cut against hard bargaining—or, because Anthropic depends on them for both capital and compute, give them practical influence the charter denies them.
• The ties have already drawn antitrust scrutiny and could draw more.

There is a disclosure asymmetry that matters for public investors. Analyses of the S-1 note that Anthropic defines related parties by holdings of voting stock. Amazon and Google hold non-voting Class N, so the two partners with the largest stakes and the largest contracts are not in the formal related-party financing tables. Their post-IPO share counts are blank. The economic conflict is discussed at length in risk factors; it is not presented as a classic related-party transaction schedule with named percentages.

Anthropic’s own answer, repeated across years, is that dual (now multi) sourcing is what preserves independence. In a May 2024 Bloomberg interview, CEO Dario Amodei argued that partnering with both Google and Amazon produces more independence than reliance on a single hyperscaler. A February 2025 court submission in the Google antitrust case, filed on Anthropic’s side of the record, states that Google’s investments “give it no role in Anthropic’s governance and no control over its operations,” that Google has “no voting rights, board seats, or even board observer rights,” and no exclusive product rights—and that buying compute from multiple partners is how Anthropic avoids dependence on any one supplier. Amazon, when it took its second $4 billion tranche to $8 billion in November 2024, was described by Anthropic and by CNBC as remaining a minority investor with no board seat. The UK CMA reached a similar conclusion on the earlier Google stake: non-voting, no board seat, not a relevant merger situation.

That legal position is accurate on corporate formalities. It does not answer the commercial question the S-1 itself flags. A supplier that can withhold or reprice the capacity Claude is trained and served on does not need a board seat to affect strategy. The minimum-spend clauses make the dependence durable: Anthropic pays the Google and Amazon floors even if it shifts workloads elsewhere, which raises the cost of actually using the multi-cloud option the independence argument depends on.

What a competitor or public shareholder should take from this

Formal control is locked to the Founder LLC and the Long-Term Benefit Trust for the foreseeable future. Amazon and Google cannot elect directors, cannot outvote the Class F share, and cannot convert their own Class N into votes without selling it away. Caps at 12% (Amazon) and 14.99% (Google), both waivable by the holder, were built to stay under control and merger-review lines while still allowing very large economic exposure—Amazon alone marked its position at roughly $190 billion at June 30, 2026.

The contestable ground is capacity, distribution, and information, not the proxy card. Nearly half of recent revenue already clears through partners who also field competing models and who are contractually owed more than $224 billion of minimum cloud spend between them, before Broadcom TPU leases and other suppliers. The S-1 treats that overlap as a risk to compute access and to alignment of incentives, while keeping the same parties outside the voting-stock definition of related parties. For anyone underwriting ANT, the governance story and the commercial story point in opposite directions: the charter says neither Amazon nor Google can control the company; the contracts say both can make independence expensive.

---

Recent Findings Supplement (November 2026)

Anthropic’s October 26, 2026 S-1 IPO filing (with a November 2 amendment) provides the first detailed public accounting of its ties to Amazon and Google, confirming large non-voting stakes, strict ownership caps, zero board or observer rights, and explicit statements that neither investor can control the company.[1][2]

This filing, released just weeks before expected IPO marketing (potentially the week of November 9), adds fresh quantitative data on investments, compute commitments, revenue routing, and risk factors that were previously estimated or confidential. It underscores Anthropic’s structure as a public benefit corporation designed to insulate mission-driven governance from investor influence.[3]

Ownership Stakes and Recent Investments

The S-1 details Amazon’s and Google’s equity positions as primarily non-voting preferred stock and convertible notes, with explicit post-conversion caps that prevent control. Amazon’s total investment reaches approximately $18 billion (including ~$8 billion in notes from 2023–2025 and $5 billion each in Series G and H preferred in Q2 2026), while Google committed up to $40 billion ($10 billion upfront in March 2026 Series G-2 plus performance-linked amounts).[4][5]

• Amazon holds $6.2 billion in outstanding convertible notes as of June 30, 2026 (valued at $97.9 billion by Amazon; its overall Anthropic holdings carried at $190.4 billion). These convert into up to ~1.84 billion non-voting Class N shares, subject to a 12% ownership cap of Anthropic’s partially diluted share count (waivable by Amazon). Excess notes remain outstanding and can convert later if room opens under the cap. Post-IPO estimates range from 15–21% or roughly one-fifth if fully converted before the cap binds.[4][6]
• Google’s stake stood at ~14% per 2025 court filings (capped at 15% under its agreement, no voting rights or board seat). It added $10 billion in March 2026 and another $10 billion in May 2026; estimates of its current stake post-dilution range from ~7–8% to 10–15%. It holds an option for up to $10 billion in Class N shares at the IPO price and can draw up to $20 billion more under a March 2026 note facility through 2030 (performance-tied). Ownership capped at 14.99% (waivable).[1][2]
• Both convert primarily into Class N common stock post-IPO, which carries identical economic rights to Class A but no voting rights (except as required by law) until sold to an unaffiliated third party.[7]

These figures represent new, filing-backed precision beyond prior estimates and highlight how the stakes function more as commercial alignments than governance tools.

For competitors or new entrants: The caps and non-voting structure show Anthropic successfully attracted massive strategic capital while retaining founder control, but they also mean Amazon and Google have skin in the game that could influence pricing or capacity negotiations in their favor as suppliers.

Governance Limits, Board Rights, and Control Statements

The S-1 confirms Amazon and Google have no board seats or observer rights. Anthropic explicitly structures governance to prevent external control, with the seven co-founders holding 50.1% voting power via a single Class F share (through Founder LLC, no economic rights) and a Long-Term Benefit Trust electing a board majority via Class T shares.[7][5]

• Ownership caps (Amazon 12%, Google 14.99%) and Class N non-voting status directly limit influence; the S-1 notes these holdings represent a “significant economic interest” but stresses the structure ensures founders and the trust steer the company.[2]
• Anthropic describes its Public Benefit Corporation setup and founder supervoting as protecting its mission; the filing states that relationships with investors like Amazon and Google do not confer control.[7]
• The trust (independent trustees with no financial stake in Anthropic) elects a majority of directors; founders retain influence over key matters even after potential sunset of certain structures.[3]

This adds concrete governance details absent from earlier reporting.

Implications: Entrants cannot rely on equity stakes alone for influence; Anthropic’s model demonstrates how AI labs can raise billions from rivals while embedding mission-protective governance that sidelines investor voting power.

Commercial Agreements: Cloud, Chips, and Distribution

The filing quantifies minimum multi-year compute commitments (often payable regardless of usage) and revenue concentration through the partners’ marketplaces.[8]

• Amazon: Minimum $110 billion AWS spend commitment through April 2036 (10-year horizon cited in some reports). Includes Trainium chips and Bedrock distribution.[4]
• Google: Minimum $114.5 billion Google Cloud spend through July 2033; additional TPU capacity deals. Sells Claude via Vertex AI.[8]
• Combined with Microsoft ($44.9 billion): At least $269 billion in Big Tech compute commitments. Additional Broadcom chip leases (~$161 billion) and others push total infrastructure obligations much higher.[9]
• Distribution: 47–48% of revenue routed through Amazon and Google cloud marketplaces (e.g., ~$2.16 billion or 47% in 2025; ~48% in H1 2026), with ~16–17% fees retained by the platforms. These partners also collect a growing share of receivables.[10]

Spending on compute/infrastructure reached nearly $9 billion in H1 2026 alone.[6]

For market participants: These take-or-pay-style deals lock in revenue for the hyperscalers while creating deep interdependence; new cloud or chip providers face high barriers as Anthropic diversifies but remains heavily committed.

Conflicts of Interest and Related-Party Risk in the IPO Filing

The S-1 devotes substantial space (dozens of pages of risk factors) to the “complex dynamics” of these relationships, describing Amazon, Google, and Microsoft as simultaneous “investors, customers, distribution partners, and competitors.”[11]

• Key risks flagged: Potential conflicts where partners could favor their own models (e.g., Gemini), gain visibility into Anthropic’s pricing/terms, curtail compute access, or trigger further antitrust scrutiny. Dependence on a limited number of suppliers “could adversely affect our access to compute.”[10][12]
• Related-party transactions are detailed at length, including the equity investments, notes, and commercial deals. The filing notes these create “potential conflicts of interest that are difficult to fully anticipate or mitigate.”[11]
• Broader context: Anthropic also highlights AI safety/existential risks and its governance structure’s limits on ordinary shareholder influence.

This level of explicit, quantified disclosure on frenemy dynamics is a notable addition from the filing.

Competitive takeaway: The risks section serves as a blueprint for how dependent AI labs must navigate supplier-investor-competitor overlaps; pure-play entrants may face fewer conflicts but also less access to capital and distribution.

Overall, the S-1 transforms prior estimates into binding, public commitments and risk disclosures just ahead of Anthropic’s planned listing, highlighting both the scale of Big Tech entanglement and the deliberate governance firewalls Anthropic has erected. No major updates appear between the initial filing and early November 2026 reports.